The Gilded Ace: Luck, Location, or a Leak?

Inspiration

Casino performance is influenced by more than how much money machines generate. Machine reliability, customer traffic, location, and operating conditions all contribute to overall performance. We wanted to explore how data analytics could help casino operators distinguish between changes in demand, machine performance, and operational inefficiencies.

Our goal was to build a tool that turns raw operational data into insights that could support better business decisions.

What It Does

The Gilded Ace is an interactive analytics dashboard designed to help casino operators understand performance across 310 gaming machines over a six-month period.

The dashboard brings together five areas of analysis:

Executive Overview: Summarizes key performance indicators and overall operational trends.
Demand & Traffic: Examines traffic patterns by day, time, and location to identify differences in customer activity.
Machine Performance: Compares machine productivity, configuration-adjusted performance, and maintenance costs to identify machines that may need attention.
Guests & Loyalty: Explores guest activity and loyalty patterns.
Forecast & Recommendations: Uses historical performance to forecast demand and translates findings into a proposed operational budget.

We used Python for data processing and analysis, Pandas for data manipulation, Plotly for interactive visualizations, and Streamlit to build the dashboard.

Our analysis examined machine-level performance, daily demand, occupancy, maintenance records, and guest activity. We also developed a forecasting approach using historical data, comparing baseline and trend-based methods. To evaluate forecasting performance, we trained on January through April and tested on May and June, preserving chronological order.

Key Findings

Our analysis revealed several operational patterns:

June coin-in per active machine-day was approximately 8.7% lower than February, while the hold percentage remained relatively stable.
Saturday had the strongest demand, while Tuesday had the weakest.
The Back Forty zone showed comparatively low occupancy and productivity.
Ridgeline machines had a higher observed maintenance frequency than the other manufacturers in the dataset.

For forecasting, our weekday-average baseline achieved a 5.07% WAPE on weekday-level averages, compared with 7.13% for the weekday-and-trend regression model. This reinforced the importance of evaluating simpler models rather than automatically choosing more complex ones.

One of our biggest challenges was turning multiple datasets into a consistent analytical workflow while accounting for missing values, inactive machines, different machine configurations, and varying levels of data granularity.

We learned that comparisons must account for operating exposure and machine configuration to avoid misleading conclusions. We also learned that correlation does not establish causation: an observed difference in machine performance does not automatically explain why it occurred.

Ultimately, this project showed us how data engineering, exploratory analysis, forecasting, and visualization can work together to transform operational data into practical recommendations.